Abstract
Audio Fingerprinting (AFP) is a fast way of music retrieval. It first records a segment of a music through the microphone on a cellphone or tablet device, and sends the recorded segment to the server for AFP computation. The server returns the most possible song to the user. However, in a real life scenario, a user commonly records the sound in a noisy environment, such as a restaurant or a supermarket. The noise might distort the recording and thus degrades the accuracy of AFP. The goal of my research is to improve the accuracy of the system in a noisy environment. The recognition system was developed in two stages. The first stage compute the confidence score for the query. The query with a low confidence score goes to the second stage for re-ranking. In the second stage, the frequency and time between the query and top 10 songs obtained from the first stage are compared, and the top 10 songs are re-ranked to improve the recognition accuracy. Three learning to rank methods are used to deal with the ranking problem, including the pointwise, the pairwise and the listwise approaches. Experimental result shows that the proposed re-ranking method is able to improve the recognition rate.